AmenAdem/Tunisian-Economy-Knowledge-Graph

Tunisian Economy Knowledge Graph # Tunisian Economy Knowledge Graph A knowledge graph system that extracts and visualizes relationships between companies, people, and organizations in the Tunisian economy using natural language processing and graph databases. ## Overview This project automatically processes documents (PDFs, web articles) about Tunisian businesses and creates a queryable knowledge graph showing ownership structures, board memberships, and business relationships. It handles multilingual content (French, Arabic, English) and provides an interactive visualization interface. ## Features - **📄 Document Processing**: Upload PDFs or scrape Tunisian news sources (TAP, African Manager, IlBoursa) - **🔍 Entity Extraction**: Hybrid NLP pipeline using spaCy and local LLM (Ollama) to identify companies, people, and organizations - **🧠 Multilingual Support**: Handles Arabic, French, and English business names with fuzzy matching and entity resolution - **📊 Knowledge Graph**: Neo4j graph database with full traceability to source documents - **🔎 Natural Language Queries**: Ask questions like "Who owns Poulina Group?" and get graph-based answers - **🌐 Interactive Visualization**: React-based graph explorer using Cytoscape.js ## Architecture ### Tech Stack - **Backend**: Python 3.11+, FastAPI, spaCy, PyMuPDF - **Graph Database**: Neo4j 5.x - **Frontend**: React 18, Vite, Cytoscape.js - **NLP**: spaCy (French/English models), Ollama (local LLM for relationship extraction) ### 5-Layer Pipeline ``` ┌─────────────────┐ │ 1. Ingestion │ PDF upload, web scraping └────────┬────────┘ │ ┌────────▼────────┐ │ 2. Processing │ Text extraction, OCR, chunking └────────┬────────┘ │ ┌────────▼────────┐ │ 3. Extraction │ NER (entities) + LLM (relationships) └────────┬────────┘ │ ┌────────▼────────┐ │ 4. Resolution │ Entity deduplication, multilingual matching └────────┬────────┘ │ ┌────────▼────────┐ │ 5. Storage │ Neo4j graph with traceability └─────────────────┘ ``` ## Quick Start ### …